Comparing analytical strategies for balancing site-level characteristics in stepped-wedge cluster randomized trials:...

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Epidemiology

Background:

• Stepped-wedge cluster randomized trials (SWCRTs) sequentially randomize clusters to interventions.
• Sequential imbalances in cluster characteristics can bias SWCRT estimation.
• This study investigates the impact of balancing cluster-level characteristics in SWCRTs.

Purpose of the Study:

• To examine the effects of balancing cluster-level characteristics in SWCRTs.
• To quantify cluster-level imbalance using a novel index.
• To assess the impact of imbalances on trial efficiency under various scenarios.

Main Methods:

• Developed a novel imbalance index based on Spearman correlation and rank regression.
• Conducted a simulation study varying number of sites, sample size, crossover timepoints, ICC, and effect sizes.
• Assessed both constant and gradual 'learning' treatment effects, measuring efficiency via RRMSE and relative mean bias.

Main Results:

• Fully-balanced designs consistently showed the highest efficiency (lowest RRMSE), particularly with learning effects.
• Increasing imbalance led to a decreasing trend in efficiency.
• Efficiency loss ranged from 52.5% to 191.9% in a 12-site example, with improvements seen for larger sample sizes, more sites, smaller ICC, and larger effect sizes.

Conclusions:

• Pre-balancing cluster characteristics significantly enhances efficiency in SWCRTs, especially when treatment effects exhibit a learning curve.
• The benefits of pre-balancing are most pronounced when accounting for dynamic treatment effects.
• This highlights the importance of pre-balancing for robust estimation in SWCRTs with time-varying effects.